How big of an area is baked lighting meant for?

Is there a size of a level that baked lights should be used for? Feels sometimes like your baked lighting can’t cover the size of a large room say before memory, space and quality becomes an issue when creating light maps. Obviously an open world space can’t be baked right?

I’m creating a space station interior. The map is all interiors with lots of rooms and corridors. It’s meant to be one space with no load times. So not open world but not a small place by any means. Is it bakable? How does lightmapping work when paired with level streaming?

I have no idea how memory allocation works for lightmaps, but you can bake lightmaps to a virtual texture so I assume at least in that case it must not require the entire world’s texture to be loaded into memory at all times. (currently virtual texture baking has a bug though)

Lightmaps are routinely used for large levels, in the Borderlands 3 stream they noted they used lightmaps to light their levels. I think the issue with open worlds may have more to do with bake times and the fact that openworld games have a dynamic day/night cycle more often than not. In Borderlands 3’s case they just invented lore to explain why the shadows never move.

Just remember that each mesh gets processed on a single thread so modular pieces will bake much faster than a single mesh with an equivalent lightmap resolution.

Plenty of large, and some open world ish games use baked indirect lighting and dynamic shadows. Last of Us 2 and Gears of War 5 come to mind.

So how does it work for Unreal. Do I just keep developing my level and every time I change something I just rebake everything?

Greybox a level and rough in the lighting. Rebuild the lighting whenever large chunks are done/updated. You can bake lighting pretty quick if you’re only using it for indirect lighting.

Volumetric Lightmaps also give much better previews of what baked lighting will look like, much better than how it worked with indirect lighting cache.

As pointed out, most of the issue with baked lighting for large scale games is the baking time, but you also need to factor in memory. Virtual Texturing can help with the streaming portion of many lightmaps, but it doesn’t impact the storage size. This is why most of the larger games have low quality bakes mostly capturing bounce lighting and not the subtle indirect shadows - they aren’t doing 1k lightmaps everywhere. If you can speed up the bake time with a Swarm network, you can get decent results with relatively low lightmap resolutions but extremely high bake settings.

Your limitation for lightmaps is mostly up to the computer you’re building lighting on, in terms of memory and computing power. Once lightmaps are built, the engine has techniques for optimizing memory usage since it doesn’t need all of the image maps loaded at full quality. Still, you keep the lightmap resolution as low as you’re OK with and go from there.

As you’re developing, you would start very simple and if you need to see what the lighting is like you do a draft build and you try to avoid doing a full quality lightmap bake until the end of development.

There’s a lot to lightmapping I need to learn about after reading these replies. What about level streaming? Does each level have a separate ‘bake’? Or is streaming and light baking not a good combination?

Why is light baking seem to be the optimal type for the game? Dynamic lighting, or semi-dynamic (stationary lights and mesh distance fields, etc), is rather comprehensive.